Augment AI improves the robustness of machine learning models in industrial manufacturing by generating synthetic training data, addressing data imbalances, and systematically optimizing models and training processes.
Target users
- Companies in manufacturing and engineering looking to improve the accuracy and reliability of their AI models
- Companies with existing datasets that lack sufficient examples of rare defects, failures, or critical operating conditions
- AI developers and data scientists working with time-series, tabular, or image data
Description
Industrial production environments generate large amounts of data during normal operation, while representative examples of defects and failures are often scarce. This imbalance can limit the accuracy, robustness, and generalizability of machine learning models.
Augment AI addresses this challenge by generating additional synthetic training data while preserving the statistical structures and physical properties of the original datasets. The service is based on AutoAug, a framework developed at the Karlsruhe Institute of Technology that automatically selects and optimizes data augmentation strategies specifically targeted at time series and transient data. HammerHAI can further provide state-of-the-art mechanisms for different data sets.
Depending on the data and the application, Augment AI combines established augmentation methods with advanced generative approaches, including diffusion and continuous flow models. In addition, it uses automated hyperparameter optimization methods such as Optuna, Hyperopt, and Bayesian optimization to jointly improve augmentation strategies, model configurations, and training processes.
Augment AI supports the following data types:
Time-series data
- Apply sensor-specific transformations such as jittering, scaling, and time warping.
- Automatically identify suitable enhancement strategies while preserving the meaning and characteristics of the original signals.
Tabular data
- Generate additional samples for under-represented classes using established techniques such as SMOTE and ADASYN.
- Preserve relevant statistical relationships through controlled interpolation in the feature space.
Image data
- Generate additional variations of existing defects using geometric and structural transformations.
- Preserve essential defect characteristics while increasing the diversity of training examples.
Applications
Augment AI is particularly valuable for industrial AI applications where data imbalances or limited training examples affect model performance. Typical use cases include:
- Robust defect detection: Improve defect classification in highly imbalanced datasets, such as production data containing 99% nondefective and 1% defective products.
- Predictive maintenance: Generate additional examples of rare but critical failure scenarios using existing machine and sensor data.
- Model validation and optimization: Introduce synthetic edge cases and systematically optimize model parameters to improve stability, accuracy, and generalization.
How it works
The Augment AI service combines data augmentation and model optimization in four steps:
Data upload
- Provide existing datasets, such as CSV files containing tabular or time-series data, or image archives.
- Optionally, provide an existing machine learning model architecture.
Data analysis
- Analyse class distributions, data types, and sensor characteristics.
- Identify suitable augmentation approaches and define the search space for augmentation and model hyperparameters.
Augmentation and Optimization
- Generate synthetic training examples using data-specific augmentation strategies.
- Automatically optimize model parameters and training configurations to make effective use of the augmented datasets.
Validation and Results
- Receive augmented datasets together with a validation report.
- Obtain the trained and optimized model, including performance metrics and documented improvements.
How to get access
Augment AI is currently under development and is available upon individual request.
Via Email: Please send a brief description of your application and available datasets to access@hammerhai.eu.
Service provided by
Contact
Disclaimer
Augment AI is currently in its beta phase and is being developed. Performance in production environments cannot yet be guaranteed. Deployments in production or production-like environments require individual consultation and coordination.
